Hi. Is there information regarding the consumption of Primary Engine or APE resources based upon Polling method? I would expect the ICMP-only uses less resources than does SNMP. However, when it comes to SAM... is Agent or WMI a bigger consumer of poller resources? And how about comparing Agent vs SNMP for SNMP-capable monitor elements? Would installing an agent, AND removing SNMP, give better performance versus using SNMP only, or versus using SNMP and WMI?
Any data on this?
Agent is supposed to be lower overhead than both WMI and SNMP, and SNMP is lower overhead than WMI. The agent initiates all of it's data collection locally based on it's own timer instead of your poller having to do the work of establishing a connection (Only for WMI, SNMP is using UDP), then requesting metrics, the node processing the request, sending requested data back across the network with the WMI/SNMP overhead, poller receives the response, breaks it down and parses out the juicy bits to put in the DB.
Compare that with how the agent locally queries the node, parses the info, bundles it up for maximum transmission efficiency and sends just the juicy bits across, with the perk of encrypting it all for you unlike snmpv2.
I know I've seen some benchmarking done on here and there was a small improvement in reduced network traffic and cpu load for the poller and IIRC the monitored node as well, but the impact of polling on each individual monitored node is pretty trivial unless you are doing something really exotic.
For places with SAM I still prefer the ease of access and administration I get from using WMI, despite the performance penalty compared to SNMP or the agent. Keeping track of the agents and deploying them, updating them, restarting them when they hang, etc is just that much more administrative overhead for me and I've never worked in a place where squeezing 1% more efficiency from the nodes was worth the added payroll it cost for me to keep on top of that stuff. Being able to build all kinds of advanced templates and posh scripts and everything else in SAM and just tell everything to inherit the credentials from the node is a real quality of life improvement for me.
On the other hand, with Linux servers I strongly prefer the agent versus SNMP since net-snmp has a bag of quirks and limitations. Linux systems also tend to be more likely to already be using automation and management tools like puppet/etc so we can leverage those to distribute the agents to new builds pretty easily, the Linux agent seems a bit more solid in terms of not randomly losing the connection, and I just generally don't have to spend much time worrying about them.
It as well comes with certain pros and cons, there are few cool posts regarding the same worth looking into them
And finally a Geek Speak article on Agent vs Agentless
Hope it helps
SolarWinds solutions are rooted in our deep connection to our user base in the THWACK® online community. More than 150,000 members are here to solve problems, share technology and best practices, and directly contribute to our product development process. Learn more today by joining now.